Group Decision Analysis Algorithms with EDAS for Interval ...algorithms of the EDAS method for group multi-criteria decision making with fuzzy sets. In the first proposed EDAS extension
Post on 12-Oct-2020
6 Views
Preview:
Transcript
51
BULGARIAN ACADEMY OF SCIENCES
CYBERNETICS AND INFORMATION TECHNOLOGIES Volume 18, No 2
Sofia 2018 Print ISSN: 1311-9702; Online ISSN: 1314-4081
DOI: 10.2478/cait-2018-0027
Group Decision Analysis Algorithms with EDAS for Interval
Fuzzy Sets
Galina Ilieva
University of Plovdiv Paisii Hilendarski, 4000 Plovdiv, Bulgaria
E-mail: galili@uni-plovdiv.bg
Abstract: The purpose of this paper is proposing, analyzing and assessing two new
algorithms of the EDAS method for group multi-criteria decision making with fuzzy
sets. In the first proposed EDAS extension for distance measure between two interval
Type-2 fuzzy numbers is applied Graded Mean Integration Representation (GMIR).
The second algorithm takes into account the proximity between the fuzzy alternatives
and its similarity measure is Map Distance Operator (MDO). The two new
algorithms are verified by a numerical example. Comparative analysis of obtained
rankings demonstrates that GMIR extension is more reliable as an interval Type-2
fuzzy alternative to Evaluation based on Distance from Average Solution (EDAS). In
case that time is of the essence, the MDO EDAS could be preferred.
Keywords: Multi-criteria group decision making, fuzzy decision making, Interval
Type-2 fuzzy sets, EDAS.
1. Introduction
The development of modern organizations can be described by uncertainty and it
depends on a number of factors, related to preferred business model and surrounding
environment. Vagueness that accompanies organizations’ activities, impedes
conducting precise experiments and leads to an inability to do exact calculations.
Moreover, data necessary for making economic decisions are varied, growing rapidly
and hard to extract. Therefore, rational management of a modern organization may
be regarded as a multi-criteria task with inaccurate and incomplete information. As,
in the general case, there is no single solution to a multi-criteria task, several different
approaches for taking into account and overcoming impreciseness in management
have been proposed, among them fuzzy sets [5-7, 18, 20] and fuzzy relations [14-17].
Research continues and employs more advanced forms of classic fuzzy sets.
New mechanisms for crisp values calculating [19, 24] and new aggregation operators
for summarizing relations [4, 11, 22] are presented.
The purpose of this research is to propose and compare two Interval Type-2
Fuzzy Sets (IT2FSs) algorithms of the Evaluation based on Distance from Average
Solution (EDAS) method. The rest of the paper is organized as follows:
Contemporary IT2FSs decision-making applications and existing EDAS
52
modifications are introduced in Section 2. Section 3 defines the basic operations for
measuring distance and similarity with Interval Type-2 Fuzzy Numbers (IT2FNs) and
describes the implementation steps used in fuzzy EDAS. Section 4 introduces the
peculiarities in new fuzzy EDAS methods for multi-criteria group decision-making.
Section 5 provides a numerical example, illustrating the proposed variants. Finally,
results are compared to those obtained when applying other fuzzy EDAS and TOPSIS
modifications.
2. Literature review
Alternatives’ ranking usually requires both quantitative and qualitative evaluations.
Interval Type-2 fuzzy sets are one of the approaches that help decision makers handle
uncertainty and vagueness of data. A b d u l l a h, A d a w i y a h and K a m a l [1] have
investigated a new decision making method of Interval Type-2 Fuzzy Simple
Additive Weighting (IT2FSAW) as a tool in dealing with ambiguity and imprecision.
The new method is applied to establish a preference in ambulance location [1].
In order to ensure more effective multi-criteria decision-making in uncertain
environments, Z h o n g and Y a o [25] extend the elimination and choice translating
reality (ELECTRE) method using interval Type-2 fuzzy numbers. They propose a
α-based distance method for measuring proximity between IT2FNs. Additionally, an
entropy measure for the IT2FNs and an entropy weight model are developed to
objectively determine the criteria weights without any weight information [25].
The aim of S o n e r, C e l i k and A k y u z [21] is to provide not only a hybrid
theoretical methodology in multiple-attribute decision making problems, but also a
practical application in maritime transportation industry. The proposed hybrid
approach integrates Analytic Hierarchy Process (AHP) method into
VlseKriterijumska Optimizacijia I Kompromisno Resenje (VIKOR) technique in an
Interval Type-2 Fuzzy (IT2F) environment. While AHP and VIKOR provide a
comprehensive framework to solve Multiple-Attribute Decision Making (MADM)
problems in maritime transportation industry, interval Type-2 fuzzy sets enables
dealing with uncertainty characteristic of linguistic assessments of decision makers.
Besides its robust theoretical insight, the paper has practical contribution to the naval
engineers, classification societies and ship owners who have difficulty in deciding
appropriate hatch cover type during construction of the ship [21].
EDAS is a relatively new method of MCDM, proposed by G h o r a b a e e et al.
[10]. In EDAS, ideas from adaptive methods in MCDM TOPSIS and VIKOR have
been improved. The numerous applications of the method show its potential for
coping with various problems, for example, in warehouse management [9],
sustainable development management [23], etc.
3. Methodology
The paper gives preference to interval Type-2 fuzzy sets assessments, since, on one
hand, they are more flexible than classic Type-1 fuzzy sets in accounting for
vagueness and uncertainty in data. On the other hand, it relies on a kind of Type-2
fuzzy sets known as interval Type-2 fuzzy numbers (IT2FNs), as it is able to deal
53
more effectively with ambiguity, has better processing abilities, and simpler
computations in comparison to classic Type-2 fuzzy numbers.
Suppose that �̃̃� and �̃̃� are two trapezoidal IT2FNs, where:
�̃̃� = (�̃�T | 𝑇 ∈ {𝑈, 𝐿}) = (𝑎𝑖T; 𝐻1𝐴
T ; 𝐻2𝐴T | 𝑇 ∈ {𝑈, 𝐿}, 𝑖 = 1, 2, 3, 4),
�̃̃� = (�̃�T | 𝑇 ∈ {𝑈, 𝐿}) = (𝑏𝑖T; 𝐻1𝐵
T ; 𝐻2𝐵T | 𝑇 ∈ {𝑈, 𝐿}, 𝑖 = 1, 2, 3, 4),
and 𝑃 (�̃̃�) is a crisp number.
The basic arithmetic operations for working with IT2FNs are described in the
specialized literature and we will not comment them here.
3.1. Graded mean integration representation of IT2FNs
Let �̃�T| 𝑇 ∈ {𝑈, 𝐿}) = (𝑎𝑖T; 𝐻1𝐴
T ; 𝐻2𝐴T |𝑇 ∈ {𝑈, 𝐿}, 𝐻1𝐴
T = 𝐻2𝐴T , 𝑖 = 1, 2, 3, 4) are the upper
and the lower trapezoidal membership function of �̃̃� respectively with the given
shape function:
(1) 𝜇𝐴T =
{
(
𝑥−𝑎1T
𝑎2T−𝑎1
T)𝑛
when 𝑥 ∈ [𝑎1T, 𝑎2
T),
𝐻1𝐴T when 𝑥 ∈ [𝑎2
T, 𝑎3T],
(𝑎4T−𝑥
𝑎4T−𝑎3
T)𝑛
when 𝑥 ∈ (𝑎3T, 𝑎4
T],
0 otherwise,
where n > 0. If n = 1, then �̃�T are known as normal trapezoidal fuzzy numbers.
According to the graded mean integration representation formula [3], the crisp
value of �̃̃� can be defined by the next equation:
(2) 𝑃 (�̃̃�) =1
2(𝑃(�̃�U) + 𝑃(�̃�L))=
=1
2∑ ∫ ℎ[(𝑎1
T + (𝑎2T − 𝑎1
T)ℎ1
𝑛 + (𝑎4T − (𝑎4
T − 𝑎3T)ℎ
1
𝑛)] 𝑑ℎ/ ∫ ℎ 𝑑ℎ 𝐻𝑝𝐴T
0
𝐻𝑝𝐴T
0𝑇∈{𝑈,𝐿}, 𝑝=1,2 =
=1
2∑ ∫ [(𝑎1
T + 𝑎4T)ℎ + (𝑎2
T − 𝑎1T − 𝑎4
T + 𝑎3T)ℎ
𝑛+1
𝑛 )] 𝑑ℎ/ ∫ ℎ 𝑑ℎ𝐻𝑝𝐴T
0
𝐻𝑝𝐴T
0𝑇∈{𝑈,𝐿}, 𝑝=1,2 =
=1
2((
𝑎1U+𝑎4
U
2+
(𝑎2U−𝑎1
U−𝑎4U+𝑎3
U)𝐻1𝐴U
3) + (
a1L+a4
L
2+
(𝑎2L−𝑎1
L−a4L+a3
L)𝐻1𝐴L
3)).
3.2. Map distance between IT2FNs
The degree of similarity between two interval Type-2 trapezoidal fuzzy numbers �̃̃�
and �̃̃� based on map distance can be determined as follows [2]:
1. Computation of the distance values ∆𝑎𝑖 and ∆𝑏𝑖, i = 1, 2, 3, 4.
In case of interval Type-2 trapezoidal fuzzy numbers �̃̃� and �̃̃�, the distance
values between the upper and lower trapezoidal fuzzy numbers are calculated as
follows:
(3) ∆𝑎𝑖 = |𝑎𝑖U−𝑎𝑖
L| and ∆𝑏𝑖 = |𝑏𝑖U−𝑏𝑖
L|, 𝑖 = 1, 2, 3, 4.
2. Evaluation of the degree of similarity 𝑆(�̃�∆, �̃�∆) between ∆𝑎𝑖 and ∆𝑏𝑖.
a) Calculation of the standard deviations ∆𝑆𝑎 and ∆𝑆𝑏 between the upper and
lower fuzzy numbers:
(4) �̅�U =1
4(𝑎1
U + 𝑎2U+𝑎3
U+𝑎3U), �̅�L =
1
4(𝑎1
L + 𝑎2L+𝑎3
L+𝑎3L),
(5) 𝑆𝐴U = √∑ (𝑎𝑖
U−�̅�U)24𝑖=1
3, 𝑆𝐴L = √
∑ (𝑎𝑖L−�̅�L)24
𝑖=1
3,
54
(6) ∆𝑆𝑎 = |𝑆𝐴U − 𝑆𝐴L|.
Similarly, find �̅�U, �̅�L, 𝑆�̃�U , 𝑆�̃�L , and ∆𝑆𝑏.
b) Determination of the map distance between the upper and lower trapezoidal
fuzzy numbers:
(7) 𝑇∆ =1
2[(2 −
1+max{|∆𝑎2−∆𝑎1|,|∆𝑏2−∆𝑏1|}
1+min{|∆𝑎2−∆𝑎1|,|∆𝑏2−∆𝑏1|}) + (2 −
1+max{|∆𝑎4−∆𝑎3|,|∆𝑏4−∆𝑏3|}
1+min{|∆𝑎4−∆𝑎3|,|∆𝑏4−∆𝑏3|})].
c) Evaluation of the degree of similarity 𝑆(�̃�∆, �̃�∆) ∈ [0, 1]:
(8) 𝑆(�̃�∆, �̃�∆) = [1 −√∑ (∆𝑎𝑖−∆𝑏𝑖)
24𝑖=1
2] × [1 − √
|∆𝑆𝑎−∆𝑆𝑏|
2] × [1 −
|𝐻1𝐴L −𝐻1𝐵
L |
|𝐻1𝐴U +𝐻1𝐵
U |] × 𝑇∆.
3. Computing the degree of similarity 𝑆(�̃�U, �̃�U) between �̃�Uand �̃�U.
a) Find the map distance between the upper trapezoidal fuzzy numbers:
(9) 𝑇U =1
2[(2 −
1+max{|𝑎2𝑢−𝑎1
𝑢|,|𝑏2𝑢−𝑏1
𝑢|}
1+min{|𝑎2𝑢−𝑎1
𝑢|,|𝑏2𝑢−𝑏1
𝑢|}) + (2 −
1+max{|𝑎4𝑢−𝑎3
𝑢|,|𝑏4𝑢−𝑏3
𝑢|}
1+min{|𝑎4𝑢−𝑎3
𝑢|,|𝑏4𝑢−𝑏3
𝑢|})].
b) Calculate the degree of similarity 𝑆(�̃�𝑈, �̃�𝑈) ∈ [0,1]:
(10) 𝑆(�̃�U, �̃�U) = [1 −√∑ (𝑎𝑖
𝑢−𝑏𝑖𝑢)24
𝑖=1
2] × [1 − √
|𝑆�̃�U
−𝑆�̃�U
|
2] × [
min(𝐻1𝐴U , 𝐻1𝐵
U )
max(𝐻1𝐴U , 𝐻1𝐵
U )] × 𝑇U.
4. Evaluate the degree of similarity 𝑆(�̃̃�, �̃̃�) between the trapezoidal fuzzy
numbers �̃̃� and �̃̃�:
(11) 𝑆 (�̃̃�, �̃̃�) = 𝑆(𝐴U, �̃�U)×(1+ 𝑆(𝐴∆, �̃�∆))
2.
The greater value of 𝑆 (�̃̃�, �̃̃�) means that the similarity between �̃̃� and �̃̃� is
greater.
4. Proposed EDAS algorithms via IT2FSs
Let a MCDM problem has n alternatives (A1, …, An) and m decision criteria
(C1, …, Cm) and each alternative is assessed according to these criteria. Decision matrix
�̃̃� = [�̃̃�𝑖𝑗]𝑛𝑚 shows all values which are assigned to the alternatives for each criterion.
The related weight of each criterion is shown as �̃̃� = [�̃̃�𝑗]1𝑚, where �̃̃� and �̃̃� are
IT2FNs.
4.1. EDAS modification via IT2FSs distance measure
1. Construct the average decision matrix [�̃̃�𝑖𝑗]𝑛𝑚based on experts’ evaluations.
2. Construct the average vector of weighted coefficients [�̃̃�𝑗]1𝑚.
3. Determine the average values of assessments according to criteria 𝐴�̃̃�𝑗.
4. Calculate the matrices of PD [𝑝𝑖𝑗] and ND [�̃̃�𝑖𝑗] from average solution using GMIR.
5. Calculate the weighted sum of PD [𝑠�̃̃�𝑖]and ND [𝑠�̃̃�𝑖] respectively.
6. Determine the normalized PD [𝑛�̃̃�𝑖]and ND [𝑛�̃̃�𝑖] for each alternative using GMIR.
7. Calculate the appraisal score [𝑎𝑠�̃̃�] for each alternative.
8. Rank the alternatives according to the appraisal score using GMIR.
Fig. 1. Flowchart of the group GMIR algorithm with IT2FSs and new distance measure
55
As in classic EDAS method, in the first proposed algorithm here, a pair of fuzzy
matrices is calculated – Positive Distances (PD) from the average and the Negative
Distances (ND) from the average solution. The evaluation of alternatives is made
according to higher average normalized values of PD and ND. The novelty here is
the idea to use Graded Mean Integration Representation (GMIR) in distance
calculations (Step 4, Step 6, and Step 8). The modified algorithm is given in Fig. 1.
4.2. EDAS algorithm via IT2FSs similarity measure
In the second proposed algorithm, in Step 3, a weighted average decision matrix is
constructed. In Step 5, alternatives’ proximities to the average solution are calculated.
The novelty is that instead of computing the distance between fuzzy numbers,
similarity measure is used, which utilizes the map distance operator (Section 3.2).
Due to the symmetrical nature of the map distance formula, calculating positive and
negative proximity to the average solution is not needed. The modified algorithm is
presented in Fig. 2.
1. Construct the average decision matrix [�̃̃�𝑖𝑗]𝑛𝑚based on experts’ evaluations.
2. Construct the average vector of weighted coefficients [�̃̃�𝑗]1𝑚.
3. Calculate the weighted average decision matrix [𝑤�̃̃�𝑖𝑗]𝑛𝑚
4. Determine the weighted average values of assessments according to critera �̃̃�𝑗.
5. Calculate the similarity measures between [𝑤�̃̃�𝑖𝑗]𝑛𝑚 and �̃̃�𝑗 using map distance operator.
6. Calculate the total degree of similarity of each alternative to the ideal solution [𝑑𝑠𝑗] 1𝑚.
7. Rank the alternatives according to their total degree of similarity.
Fig. 2. Flowchart of the group MDO algorithm with IT2FSs and similarity measure
A detailed step-by-step description of new algorithms is provided in the next
section.
5. Numerical example
Let an MCDM problem have four alternatives (A1, … , A4) and seven decision criteria
(C1, … , C7), and all criteria be beneficial (maximizing). Let there be three experts, who
assess the compared alternatives using IF2FNs and assign IF2FNs weights to every
criterion.
Step 1. The decision matrix �̃̃� = [�̃̃�𝑖𝑗]47 contains averaged experts’ assessments
of each alternative according to every criterion.
Step 2. The average relative weight of each criterion is shown in the vector
�̃̃� = [�̃̃�𝑗]17.
Table 1. Decision matrix and weights of the criteria
Alternative C1 C2 C3 C4 C5 C6 C7
A1 L AH VH VH H AH VH
A2 L M L L VL AH VH
A3 M M L H L M H
A4 H H L AH H H L
W H M L VH M VH AH
56
The values in the decision matrix and weighted coefficients are shown in
Table 1 as nine grade linguistic variables. For transforming every linguistic variable
into its corresponding symmetric trapezoidal IT2FN, we apply a correspondence
table (Table 2). Obtained results are given in Table 3 and Table 4 in the Appendix.
Table 2. Linguistic terms and their corresponding interval type-2 fuzzy numbers
Linguistic terms Trapezoidal IT2FNs
No influence (No) ((0,0,0.05,0.15;1,1),(0,0,0.035,0.125;0.9,0.9))
Very Low (VL) ((0,0.1,0.15,0.25;1,1),(0.025,0.115,0.135,0.225;0.9,0.9))
Low (L) ((0.125,0.225,0.275,0.375;1,1),(0.15,0.24,0.26,0.35;0.9,0.9))
Medium Low (ML) ((0.25,0.35,0.4,0.5;1,1),(0.275,0.365,0.385,0.475;0.9,0.9))
Medium (M) ((0.375,0.475,0.525,0.625;1,1),(0.4,0.49,0.51,0.6;0.9,0.9))
Medium High (MH) ((0.5,0.6,0.65,0.75;1,1),(0.525,0.615,0.635,0.725;0.9,0.9))
High (H) ((0.625,0.725,0.775,0.875;1,1),(0.65,0.74,0.76,0.85;0.9,0.9))
Very High (VH) ((0.75,0.85,0.9,1;1,1),(0.775,0.865,0.885,0.975;0.9,0.9))
Absolutely High (AH) ((0.875,0.975,1,1;1,1),(0.9,0.99,1,1;0.9,0.9))
5.1. EDAS algorithm via IT2FNs distance measure
Step 3. The matrix of average values of assessments according to each criterion
𝐴�̃̃�𝑗 is built by using the results of Step 1 and the next equation:
(12) 𝐴�̃̃�𝑗 =∑ 𝑥𝑖𝑗𝑛𝑖=1
𝑛.
Step 4. Based on Table 3, matrix of average solution from Step 3 and Equations
(13) to (16), the Positive Distances (PD) and Negative Distances (ND) matrices are
calculated:
(13) PD = [𝑝�̃̃�𝑖𝑗]4×7,
(14) ND = [𝑛�̃̃�𝑖𝑗]4×7,
(15) 𝑝�̃̃�𝑖𝑗 = {
𝜓(𝑥𝑖𝑗⊝𝑎�̃̃�𝑗)
𝑃(𝑎�̃̃�𝑗) if 𝑗 ∈ 𝐵,
𝜓(𝑎�̃̃�𝑗⊝𝑥𝑖𝑗)
𝑃(𝑎�̃̃�𝑗) if 𝑗 ∈ 𝐶,
(16) 𝑛�̃̃�𝑖𝑗 = {
𝜓(𝑎�̃̃�𝑗⊝𝑥𝑖𝑗)
𝑃(𝑎�̃̃�𝑗) if 𝑗 ∈ 𝐵,
𝜓(𝑥𝑖𝑗⊝𝑎�̃̃�𝑗)
𝑃(𝑎�̃̃�𝑗) if 𝑗 ∈ 𝐶,
where function 𝜓 is defined to determine the maximum of interval Type-2 fuzzy
number and zero as follows:
(17) 𝜓(�̃̃�) = {�̃̃� if 𝑃 (�̃̃�) > 0,
0̃̃ if 𝑃 (�̃̃�) ≤ 0,
where 0̃̃=((0, 0, 0, 0; 1, 1),(0, 0, 0, 0; 1, 1)) and B signifies the set of maximizing
criteria, and C denotes the group of minimizing criteria [9].
Steps 5-7. By using matrices of positive and negative distances and the next
Equations (18)-(22), the weighted sum of positive and negative distances (𝑠�̃��̃� and
𝑠�̃��̃�), their normalized values (𝑛�̃��̃� and 𝑛�̃��̃�), and the appraisal scores are calculated for
all alternatives:
(18) 𝑠�̃��̃� =⊕𝑗=1𝑚 (�̃̃�𝑗⊗𝑝�̃̃�𝑖𝑗),
57
(19) 𝑠�̃��̃� =⊕𝑗=1𝑚 (�̃̃�𝑗⊗𝑛�̃̃�𝑖𝑗),
(20) 𝑛�̃��̃� =𝑠�̃�1̃
max𝑖 (𝑃(𝑠�̃��̃�),
(21) 𝑛�̃�𝑖̃ = 1 −𝑠�̃�1̃
max𝑖 (𝑃(𝑠�̃��̃�),
(22) ℎ̃𝑖̃ =
𝑃(𝑛�̃̃�𝑖)+𝑃(𝑛�̃̃�𝑖)
2
Step 8. According to Table 6 (see Appendix), the ranking values of Appraisal
Scores (AS) can be calculated. The results are as follows: A1 (0.894) ≻ A4 (0.663) ≻
A3 (0.318) ≻ A2 (0.153). Therefore, A1 is the best alternative according to the given
seven criteria.
5.2. EDAS algorithm via IT2FSs similarity measure
As Step 1 and Step 2 are shared by both two algorithms, calculations continue from
the next step:
Step 3. Calculate the weighted average decision matrix [𝑤�̃̃�𝑖𝑗]47, where
𝑤�̃̃�𝑖𝑗 = w̃̃𝑗 ⊗ �̃̃�𝑖𝑗, 1 ≤ 𝑖 ≤ 4 and 1 ≤ 𝑗 ≤ 7 (Table 7, see Appendix).
Step 4. Determine the weighted average values of assessments according to
criteria [�̃̃�𝑗]17, where �̃̃�𝑗 = w̃̃𝑗⊗𝑎�̃̃�𝑗, 1 ≤ 𝑗 ≤ 7 (Table 8, see Appendix).
Step 5 and 6. Calculate the similarity measures between [𝑤�̃̃�𝑖𝑗]𝑛𝑚 and �̃̃�𝑗 using
the map distance operator. Results are shown in Table 9 (see Appendix), (Equations
(4), (5) and (6) for 𝑤�̃̃�𝑖𝑗), Table 10 (see Appendix), (Equations (4), (5) and (6) for �̃̃�𝑗)
and Table 11 (see Appendix), (Equations (7), (8), (9), (10) and (11) for 𝑤�̃̃�𝑖𝑗 and �̃̃�𝑗).
After computing the total degree of similarity of each alternative to the ideal solution
[𝑑𝑠𝑗] 1𝑚 we have: A1 ≈ 4.613, A2 ≈ 4.548, A3 ≈ 5.082, A4 ≈ 4.465. Final ranking is
as follows: A3 ≻ A1 ≻ A2 ≻ A4.
5.3. Comparison of obtained results with other heuristic EDAS methods
In order to validate the results, we solve the same task with heuristic formulae from
our literature review for determining distances between IT2FNs, such as DTraT [8]
и 𝔖 [9]. Obtained rankings are shown in Table 3. To evaluate the qualities of the
solutions obtained, we compute generalized ranking with Borda count and
Copeland’s methods. It turns out that the ranking given by the first variant is
practically identical to the two benchmarking-rankings (for Borda count A1 ≻ A4 ≻
A3 ≻ A2, for Copeland’s method A1 ≻ A4 ≻ A3 ≈ A2). This is not the case with the
map distance variant, however. Here, there are considerable shifts and the ranking is
as follows: A2 ≻ A3 ≻ A1 ≻ A4 (Table 3).
Table 3. Comparing different EDAS modifications and the corresponding Borda count and
Copeland’s Method’s (CM) ranking
Alterna-
tive
EDAS GMIR EDAS MDO EDAS DTraT EDAS 𝔖
Values Rank Values Rank Values Rank Values Rank
A1 0.894 1 4.613 2 0.863 1 0.885 1
A2 0.153 4 4.548 3 0.086 4 0.123 4
A3 0.318 3 5.082 1 0.261 3 0.292 3
A4 0.663 2 4.465 4 0.624 2 0.646 2
58
Table 3 (c o n t i n u e d)
Alterna-
tive
TOPSIS DTraT TOPSIS GMIR Borda
count
CM
points Values Rank Values Rank
A1 0.776 1 0.7841 1 23 3
A2 0.502 3 0.4827 3 9 0.5
A3 0.459 4 0.4521 4 12 0.5
A4 0.521 2 0.5281 2 16 2
Performance analysis of the two modifications shows that the mechanisms for
determining similarity differ in their time complexities. The MDО method needs
fewer calculations. The existence of two distance matrices, however, requires
additional computations, therefore the complexity of GMIR EDAS is greater than
that of MDО algorithm. For example, in the GMIR variant, two distance matrices are
calculated and normalized (Equation (18) and Equation (19), respectively).
Aggregating the pair of matrices is also an intensive computational process
(Equations (20)-(22)). The MDО variant works with a single matrix and the number
of operations is smaller (Equations (4)-(11)). Additional optimization is possible,
since the average solution’s characteristics (Equation (4)) are calculated only once.
Due to the fewer calculations, at first glance it may seem that MDО EDAS holds
the advantage. However, this statement does not apply to algorithm’s reliability,
which is very important measure of new modifications’ qualities from a practical
point of view.
Comparative analysis with IT2FNs TOPSIS and other EDAS methods shows a
deviation of MDО EDAS ranking from the average result, calculated via statistical
methods. In practice, there is a possibility that deviations decrease if instead of
average values calculated in situ, we work with optimal values, provided by
manufacturer or experts selected in advance.
6. Conclusions
The paper proposes two algorithms of EDAS in a IT2FSs environment – one with
GMIR defuzzification and another with MD method for measuring similarity. The
advantages of the algorithms are as follows: 1) expand the applicability of EDAS in
more uncertainty environments; 2) enrich the tools for distance measure in EDAS
MCDM. Using a numerical example, it was proven that the new algorithms are
suitable for MCDM in the absence of dependencies between assessment criteria.
Obtained results were compared to results from employing existing formulae for
distance between IT2FSs from specialized literature. Analysis confirms the
efficiency of the two EDAS algorithms.
In the future, we plan on conducting more experiments with other fuzzy sets
extensions and implementing the methods of IT2FSs MCDA in various fields of
applications, such as modeling and simulating complex systems.
Acknoledgements: This research was supported by the Scientific Research Fund of the University of
Plovdiv Paisii Hilendarski as a part of project SR17 FESS 012/ 24.04.2017.
59
R e f e r e n c e s
1. A b d u l l a h, L., C. A d a w i y a h, C. K a m a l. A Decision Making Method Based on Interval
Type-2 Fuzzy Sets: An Approach for Ambulance Location Preference. – Applied Computing
and Informatics, Vol. 14, 2018, Issue 1, pp. 65-72.
2. C h e n, S.-J., Z.-Y. W a n g, W.-R. L i. Calculating the Degree of Similarity between Interval-
Valued Fuzzy Numbers Based on Map Distance. – In: Proc. of International Multi Conference
of Engineers and Computer Scientists (IMECS’13), 2013, Hong Kong.
3. C h e n, S. H., C. H. H s i e h. Representation, Ranking, Distance, and Similarity of L-R Type Fuzzy
Number and Application. – Australia Journal of Intelligent Information Processing Systems,
Vol. 6, 2000, No 4, pp. 217-229.
4. C h i a o, K. Interval Type-2 Fuzzy Sets Multiple Criteria Decision Making Based on Quantifier
Guided Aggregation Ordered Weighted Averaging Operator. – In: Proc. of Joint 17th World
Congress of International Fuzzy Systems Association and 9th International Conference on Soft
Computing and Intelligent Systems (IFSA-SCIS), 27-30 June 2017.
5 . I l i e v a, G. A Fuzzy Approach for Bidding Strategy Selection. – Cybernetics and Information
Technologies, Vol. 12, 2012, No 1, pp. 61-69.
6. I l i e v a, G. Group Decision Analysis with Interval Type-2 Fuzzy Numbers. – Cybernetics and
Information Technologies, Vol. 17, 2017, No 1, pp. 31-44.
7. I l i e v a, G. TOPSIS Modification with Interval Type-2 Fuzzy Numbers. – Cybernetics and
Information Technologies, Vol. 16, 2016, No 2, pp. 60-68.
8. K a h r a m a n, C., B. O z t a y s i, E. T u r a n o g l u. Fuzzy Analytic Hierarchy Process with Interval
Type-2 Fuzzy Sets. – Knowledge-Based Systems, Vol. 59, 2014, pp. 48-57.
9. G h o r a b a e e, K. M., M. A m i r i, E. K. Z a v a d s k a s, Z. T u r s k i s. Multi-Criteria Group
Decision-Making Using an Extended EDAS Method with Interval Type-2 Fuzzy Sets. –
Economics and Management, Vol. 20, 2017, No 1, pp. 48-68.
10. G h o r a b a e e, K. M., E. K. Z a v a d s k a s, L. O l f a t, Z. T u r s k i s. Multicriteria Inventory
Classification Using a New Method of Evaluation Based on Distance from Average Solution
(EDAS). – Informatica, Vol. 26, 2015, No 3, pp. 435-451.
11. L i a o, H., Z. X u. Multiple Criteria Decision Making with Hesitant Fuzzy Hybrid Weighted
Aggregation Operators. – In: Hesitant Fuzzy Decision Making Methodologies and
Applications, Springer, 2017.
12. P e n e v a, V., I. P o p c h e v. Fuzzy Criteria Importance Depending on Membership Degrees of
Fuzzy Relations. – Compt. Rend. Acad. bulg. Sci., Vol. 61, 2008, No 5, pp. 579-584.
13. P e n e v a, V., I. P o p c h e v. Fuzzy Criteria Importance with Weighting Functions. – Compt. Rend.
Acad. bulg. Sci., Vol. 61, 2008, No 3, pp. 293-300.
14. P e n e v a, V., I. P o p c h e v. Fuzzy Ordering on Base of Multicriteria Aggregation. – Cybernetics
and Systems, an International Journal, Vol. 29, 1998, No 6, Taylor and Francis Group, Robert
Trappl, Ed., pp. 613-623.
15. P e n e v a, V., I. P o p c h e v. Models for Decision Making by Fuzzy Relations and Fuzzy Numbers
for Criteria Evaluations. – Compt. Rend. Acad. bulg. Sci., Vol. 62, 2009, No 10,
pp. 1217-1222.
16. P e n e v a, V., I. P o p c h e v. Multicriteria Decision Making by Fuzzy Relations and Weighting
Functions for the Criteria. – Cybernetics and Information Technologies, Vol. 9, 2009, No 4,
pp. 58-71.
17. P o p c h e v, I., V. P e n e v a. An Algorithm for Comparison of Fuzzy Sets. – Fuzzy Sets and
Systems, Elsevier Science Publishers, North-Holland, Amsterdam, Vol. 60, 1993, No 1,
pp. 59-65.
18. R a d e v a, I. Multicriteria Fuzzy Sets Application in Economic Clustering Problems. – Cybernetics
and Information Technologies, Vol. 17, 2017, No 3, pp. 29-46.
19. R u n k l e r, T., S. C o u p l a n d, J. R o b e r t. Interval Type-2 Fuzzy Decision Making. –
International Journal of Approximate Reasoning, Vol. 80, 2017, pp. 217-224.
20. S h a r a f, I. TOPSIS with Similarity Measure for MADM Applied to Network Selection. –
Computational and Applied Mathematics, 2017, pp. 1-18.
60
21. S o n e r, O., E. C e l i k, E. A k y u z. Application of AHP and VIKOR Methods under Interval
Type-2 Fuzzy Environment in Maritime Transportation. – Ocean Engineering, Vol. 129, 2017,
pp. 107-116.
22. W e i, G. Pythagorean Fuzzy Interaction Aggregation Operators and Their Application to Multiple
Attribute Decision Making. – Intelligent & Fuzzy Systems, Vol. 33, 2017, No 4,
pp. 2119-2132.
23. Z a v a d s k a s, Е. К., F. C a v a l l a r o, V. P o d v e z k o, I. U b a r t e, A. K a k l a u s k a s. MCDM
Assessment of a Healthy and Safe Built Environment According to Sustainable Development
Principles: A Practical Neighborhood Approach in Vilnius. – Sustainability, Vol. 9, 2017,
No 5. 702 p.
24. Z h a n g, H., J. W a n g, X. C h e n. Interval Neutrosophic Sets and Their Application in Multicriteria
Decision Making Problems. – Hindawi the Scientific World Journal, Vol. 2014, Article ID
645953. 15 p.
25. Z h o n g, L., L. Y a o. An ELECTRE I-Based Multi-Criteria Group Decision Making Method with
Interval Type-2 Fuzzy Numbers and its Application to Supplier Selection. – Applied Soft
Computing, Vol. 57, 2017, pp. 556-576.
Appendix. Step-by-step calculations
Table 4. The average decision matrix [�̃̃�𝑖𝑗]47
�̃̃�𝑖𝑗 �̃�𝑖𝑗U �̃�𝑖𝑗
L
𝑥1U 𝑥2
U 𝑥3U 𝑥4
U 𝑥1L 𝑥2
L 𝑥3L 𝑥4
L
�̃̃�11 0.125 0.225 0.275 0.375 0.150 0.240 0.260 0.350
�̃̃�21 0.125 0.225 0.275 0.375 0.150 0.240 0.260 0.350
�̃̃�31 0.375 0.475 0.525 0.625 0.400 0.490 0.510 0.600
�̃̃�41 0.625 0.725 0.775 0.875 0.650 0.740 0.760 0.850
�̃̃�12 0.875 0.975 1.000 1.000 0.900 0.990 1.000 1.000
�̃̃�22 0.375 0.475 0.525 0.625 0.400 0.490 0.510 0.600
�̃̃�32 0.375 0.475 0.525 0.625 0.400 0.490 0.510 0.600
�̃̃�42 0.625 0.725 0.775 0.875 0.650 0.740 0.760 0.850
�̃̃�13 0.750 0.850 0.900 1.000 0.775 0.865 0.885 0.975
�̃̃�23 0.125 0.225 0.275 0.375 0.150 0.240 0.260 0.350
�̃̃�33 0.125 0.225 0.275 0.375 0.150 0.240 0.260 0.350
�̃̃�43 0.125 0.225 0.275 0.375 0.150 0.240 0.260 0.350
�̃̃�14 0.750 0.850 0.900 1.000 0.775 0.865 0.885 0.975
�̃̃�24 0.125 0.225 0.275 0.375 0.150 0.240 0.260 0.350
�̃̃�34 0.625 0.725 0.775 0.875 0.650 0.740 0.760 0.850
�̃̃�44 0.875 0.975 1.000 1.000 0.900 0.990 1.000 1.000
�̃̃�15 0.625 0.725 0.775 0.875 0.650 0.740 0.760 0.850
�̃̃�25 0.000 0.100 0.150 0.250 0.025 0.115 0.135 0.225
�̃̃�35 0.125 0.225 0.275 0.375 0.150 0.240 0.260 0.350
�̃̃�45 0.625 0.725 0.775 0.875 0.650 0.740 0.760 0.850
�̃̃�16 0.875 0.975 1.000 1.000 0.900 0.990 1.000 1.000
�̃̃�26 0.875 0.975 1.000 1.000 0.900 0.990 1.000 1.000
�̃̃�36 0.375 0.475 0.525 0.625 0.400 0.490 0.510 0.600
�̃̃�46 0.625 0.725 0.775 0.875 0.650 0.740 0.760 0.850
�̃̃�17 0.750 0.850 0.900 1.000 0.775 0.865 0.885 0.975
�̃̃�27 0.750 0.850 0.900 1.000 0.775 0.865 0.885 0.975
�̃̃�37 0.625 0.725 0.775 0.875 0.650 0.740 0.760 0.850
�̃̃�47 0.125 0.225 0.275 0.375 0.150 0.240 0.260 0.350
61
Table 5. The average weights’ coefficients [�̃̃�𝑗]17
�̃̃�𝑗
�̃�𝑖𝑗U �̃�𝑖𝑗
L
𝑤1U 𝑤2
U 𝑤3U 𝑤4
U 𝑤1L 𝑤2
L 𝑤3L 𝑤4
L
�̃̃�1 0.625 0.725 0.775 0.875 0.650 0.740 0.760 0.850
�̃̃�2 0.375 0.475 0.525 0.625 0.400 0.490 0.510 0.600
�̃̃�3 0.125 0.225 0.275 0.375 0.150 0.240 0.260 0.350
�̃̃�4 0.750 0.850 0.900 1.000 0.775 0.865 0.885 0.975
�̃̃�5 0.375 0.475 0.525 0.625 0.400 0.490 0.510 0.600
�̃̃�6 0.750 0.850 0.900 1.000 0.775 0.865 0.885 0.975
�̃̃�7 0.875 0.975 1.000 1.000 0.900 0.990 1.000 1.000
Table 6. The weighted sum of distances from the average solution, their
normalized values and corresponding appraisal scores
𝑠�̃̃�𝑖 �̃�𝑖𝑗U �̃�𝑖𝑗
L
𝑘1U 𝑘2
U 𝑘3U 𝑘4
U 𝑘1L 𝑘2
L 𝑘3L 𝑘4
L
𝑠�̃̃�1 –0.035 1.142 1.828 3.371 0.242 1.347 1.622 2.958
𝑠�̃̃�2 –0.091 0.354 0.585 1.027 0.018 0.428 0.520 0.914
𝑠�̃̃�3 –0.911 0.023 0.466 1.474 –0.664 0.150 0.327 1.208
𝑠𝑝4 0.037 0.966 1.487 2.691 0.253 1.121 1.329 2.365
𝑠�̃̃�1 –0.078 0.142 0.262 0.546 –0.015 0.177 0.225 0.470
𝑠�̃̃�2 0.023 0.736 1.140 2.108 0.211 0.903 1.076 1.918
𝑠�̃̃�3 –0.079 0.377 0.632 1.246 0.048 0.452 0.554 1.077
𝑠�̃̃�4 –0.051 0.388 0.600 1.053 0.066 0.454 0.539 0.932
𝑛�̃̃�1 –0.023 0.743 1.188 2.191 0.157 0.876 1.054 1.923
𝑛�̃̃�2 –0.059 0.230 0.380 0.668 0.012 0.278 0.338 0.594
𝑛�̃̃�3 –0.592 0.015 0.303 0.959 –0.432 0.097 0.213 0.786
𝑛�̃̃�4 0.024 0.628 0.967 1.749 0.164 0.728 0.864 1.537
𝑛�̃̃�1 0.454 0.738 0.858 1.078 0.530 0.775 0.823 1.015
𝑛�̃̃�2 –1.108 –0.140 0.264 0.977 –0.918 –0.076 0.097 0.789
𝑛�̃̃�3 –0.246 0.368 0.623 1.079 –0.077 0.446 0.548 0.952
𝑛�̃̃�4 –0.053 0.400 0.612 1.051 0.068 0.461 0.546 0.934
ℎ̃̃1 0.216 0.740 1.023 1.635 0.344 0.825 0.939 1.469
ℎ̃̃2 –0.584 0.045 0.322 0.823 –0.453 0.101 0.218 0.692
ℎ̃̃3 –0.419 0.191 0.463 1.019 –0.254 0.272 0.380 0.869
ℎ̃̃4 –0.014 0.514 0.789 1.400 0.116 0.595 0.705 1.236
62
Table 7. The weighted average decision matrix [𝑤�̃̃�𝑖𝑗]47
𝑤�̃̃�𝑖𝑗 𝑤�̃�𝑖𝑗
U 𝑤�̃�𝑖𝑗L
𝑤𝑥1U 𝑤𝑥2
U 𝑤𝑥3U 𝑤𝑥4
U 𝑤𝑥1L 𝑤𝑥2
L 𝑤𝑥3L 𝑤𝑥4
L
𝑤�̃̃�11 0.078 0.163 0.213 0.328 0.098 0.178 0.198 0.30
𝑤�̃̃�21 0.078 0.163 0.213 0.328 0.135 0.228 0.251 0.34
𝑤�̃̃�31 0.234 0.344 0.407 0.547 0.360 0.466 0.493 0.59
𝑤�̃̃�41 0.391 0.526 0.601 0.766 0.585 0.703 0.735 0.84
𝑤�̃̃�12 0.328 0.463 0.525 0.625 0.360 0.485 0.510 0.600
𝑤�̃̃�22 0.141 0.226 0.276 0.391 0.160 0.240 0.260 0.360
𝑤�̃̃�32 0.141 0.226 0.276 0.391 0.160 0.240 0.260 0.360
𝑤�̃̃�42 0.234 0.344 0.407 0.547 0.260 0.363 0.388 0.510
𝑤�̃̃�13 0.094 0.191 0.248 0.375 0.116 0.208 0.230 0.34
𝑤�̃̃�23 0.016 0.051 0.076 0.141 0.023 0.058 0.068 0.12
𝑤�̃̃�33 0.016 0.051 0.076 0.141 0.023 0.058 0.068 0.12
𝑤�̃̃�43 0.016 0.051 0.076 0.141 0.023 0.058 0.068 0.12
𝑤�̃̃�14 0.563 0.723 0.810 1.000 0.601 0.748 0.783 0.95
𝑤�̃̃�24 0.094 0.191 0.248 0.375 0.116 0.208 0.230 0.34
𝑤�̃̃�34 0.469 0.616 0.698 0.875 0.504 0.640 0.673 0.83
𝑤�̃̃�44 0.656 0.829 0.900 1.000 0.698 0.856 0.885 0.98
𝑤�̃̃�15 0.234 0.344 0.407 0.547 0.260 0.363 0.388 0.51
𝑤�̃̃�25 0.000 0.048 0.079 0.156 0.010 0.056 0.069 0.14
𝑤�̃̃�35 0.047 0.107 0.144 0.234 0.060 0.118 0.133 0.21
𝑤�̃̃�45 0.234 0.344 0.407 0.547 0.260 0.363 0.388 0.51
𝑤�̃̃�16 0.656 0.829 0.900 1.000 0.698 0.856 0.885 0.98
𝑤�̃̃�26 0.656 0.829 0.900 1.000 0.698 0.856 0.885 0.98
𝑤�̃̃�36 0.281 0.404 0.473 0.625 0.310 0.424 0.451 0.59
𝑤�̃̃�46 0.469 0.616 0.698 0.875 0.504 0.640 0.673 0.83
𝑤�̃̃�17 0.656 0.829 0.900 1.000 0.698 0.856 0.885 0.98
𝑤�̃̃�27 0.656 0.829 0.900 1.000 0.698 0.856 0.885 0.98
𝑤�̃̃�37 0.547 0.707 0.775 0.875 0.585 0.733 0.760 0.85
𝑤�̃̃�47 0.109 0.219 0.275 0.375 0.135 0.238 0.260 0.35
Table 8. The elements of the weighted average solution matrix [�̃̃�𝑗]17
�̃̃�𝑗 �̃�𝑗U �̃�𝑗
L
𝑀1U 𝑀2
U 𝑀3U 𝑀4
U 𝑀1L 𝑀2
L 𝑀3L 𝑀4
L
�̃̃�1 0.195 0.299 0.358 0.492 0.219 0.316 0.340 0.457
�̃̃�2 0.211 0.315 0.371 0.488 0.235 0.332 0.354 0.458
�̃̃�3 0.035 0.086 0.119 0.199 0.046 0.095 0.108 0.177
�̃̃�4 0.445 0.590 0.664 0.813 0.480 0.613 0.643 0.774
�̃̃�5 0.129 0.211 0.259 0.371 0.148 0.225 0.244 0.341
�̃̃�6 0.516 0.669 0.743 0.875 0.552 0.694 0.723 0.841
�̃̃�7 0.492 0.646 0.713 0.813 0.529 0.671 0.698 0.788
63
Table 9. The map distance calculation for matrix [𝑤�̃̃�𝑖𝑗]47
𝑤�̃̃�𝑖𝑗 𝑤𝑥̅̅ ̅̅ U 𝑤𝑥̅̅ ̅̅ L 𝑆𝑊�̃�U 𝑆𝑊�̃�L ∆𝑆𝑤𝑥
𝑤�̃̃�11 0.196 0.193 0.104 0.082 0.022
𝑤�̃̃�21 0.196 0.240 0.104 0.086 0.019
𝑤�̃̃�31 0.383 0.477 0.130 0.095 0.036
𝑤�̃̃�41 0.571 0.715 0.156 0.103 0.053
𝑤�̃̃�12 0.485 0.489 0.124 0.099 0.025
𝑤�̃̃�22 0.258 0.255 0.104 0.082 0.022
𝑤�̃̃�32 0.258 0.255 0.104 0.082 0.022
𝑤�̃̃�42 0.383 0.380 0.130 0.103 0.028
𝑤�̃̃�13 0.227 0.224 0.117 0.092 0.025
𝑤�̃̃�23 0.071 0.068 0.053 0.041 0.011
𝑤�̃̃�33 0.071 0.068 0.053 0.041 0.011
𝑤�̃̃�43 0.071 0.068 0.053 0.041 0.011
𝑤�̃̃�14 0.774 0.771 0.182 0.144 0.039
𝑤�̃̃�24 0.227 0.224 0.117 0.092 0.025
𝑤�̃̃�34 0.664 0.661 0.169 0.133 0.036
𝑤�̃̃�44 0.846 0.853 0.145 0.116 0.029
𝑤�̃̃�15 0.383 0.380 0.130 0.103 0.028
𝑤�̃̃�25 0.071 0.068 0.066 0.052 0.014
𝑤�̃̃�35 0.133 0.130 0.079 0.062 0.017
𝑤�̃̃�45 0.383 0.380 0.130 0.103 0.028
𝑤�̃̃�16 0.846 0.853 0.145 0.116 0.029
𝑤�̃̃�26 0.846 0.853 0.145 0.116 0.029
𝑤�̃̃�36 0.446 0.443 0.143 0.113 0.030
𝑤�̃̃�46 0.664 0.661 0.169 0.133 0.036
𝑤�̃̃�17 0.846 0.853 0.145 0.116 0.029
𝑤�̃̃�27 0.846 0.853 0.145 0.116 0.029
𝑤�̃̃�37 0.726 0.732 0.138 0.110 0.028
𝑤�̃̃�47 0.245 0.246 0.111 0.088 0.023
Table 10. The map distance calculation for matrix [�̃̃�𝑗]17
�̃̃�𝑗 �̅�U �̅�L 𝑆�̃�U 𝑆�̃�L ∆𝑆𝑚
�̃̃�1 0.336 0.333 0.124 0.098 0.026
�̃̃�2 0.346 0.345 0.116 0.091 0.024
�̃̃�3 0.110 0.107 0.069 0.054 0.015
�̃̃�4 0.628 0.627 0.153 0.121 0.032
�̃̃�5 0.243 0.239 0.101 0.080 0.022
�̃̃�6 0.701 0.703 0.150 0.119 0.031
�̃̃�7 0.666 0.671 0.134 0.107 0.027
64
Table 11. The map distance calculation for matrices [𝑤�̃̃�𝑖𝑗]47 and [�̃̃�𝑗]17
𝑤�̃̃�𝑖𝑗 𝑇∆ 𝑆(𝑤�̃�∆, �̃�∆) 𝑇U 𝑆(𝑤�̃�U, �̃�U) 𝑆 (𝑊�̃̃�, �̃̃�)
𝑤�̃̃�11 0.998 0.949 0.983 0.684 0.667
𝑤�̃̃�21 0.997 0.905 0.983 0.684 0.652
𝑤�̃̃�31 0.986 0.844 0.994 0.804 0.742
𝑤�̃̃�41 0.972 0.746 0.972 0.583 0.509
𝑤�̃̃�12 0.996 0.970 0.978 0.707 0.697
𝑤�̃̃�22 0.999 0.964 0.990 0.752 0.738
𝑤�̃̃�32 0.999 0.964 0.990 0.752 0.738
𝑤�̃̃�42 0.998 0.954 0.987 0.780 0.762
𝑤�̃̃�13 0.995 0.915 0.956 0.636 0.609
𝑤�̃̃�23 0.999 0.954 0.985 0.774 0.756
𝑤�̃̃�33 0.999 0.954 0.985 0.774 0.756
𝑤�̃̃�43 0.999 0.954 0.985 0.774 0.756
𝑤�̃̃�14 0.997 0.934 0.975 0.657 0.635
𝑤�̃̃�24 0.997 0.930 0.969 0.452 0.436
𝑤�̃̃�34 0.998 0.951 0.986 0.775 0.756
𝑤�̃̃�44 0.995 0.949 0.966 0.635 0.619
𝑤�̃̃�15 0.997 0.936 0.974 0.661 0.640
𝑤�̃̃�25 0.997 0.929 0.968 0.623 0.601
𝑤�̃̃�35 0.998 0.945 0.980 0.700 0.681
𝑤�̃̃�45 0.997 0.936 0.974 0.661 0.640
𝑤�̃̃�16 0.997 0.960 0.977 0.713 0.699
𝑤�̃̃�26 0.997 0.960 0.977 0.713 0.699
𝑤�̃̃�36 0.997 0.972 0.977 0.618 0.609
𝑤�̃̃�46 0.997 0.942 0.977 0.760 0.737
𝑤�̃̃�17 0.999 0.965 0.992 0.679 0.667
𝑤�̃̃�27 0.999 0.965 0.992 0.679 0.667
𝑤�̃̃�37 1.000 0.980 0.997 0.808 0.801
𝑤�̃̃�47 0.998 0.943 0.980 0.455 0.442
Received 22.03.2018; Accepted 05.04.2018
top related